At the AAA meeting in DC, I attended a presidential address by Ray Ball and Phil Brown regarding their seminal research paper (JAR 1968).  They described the motivation for their study as a test of existing scholarly research that painted a dim picture of reported earnings. The earlier writers noted that earnings were based on old information (historical cost) or, worse yet, a mix of old and new information (mixed attributes).  The early articles concluded that earnings could not be informative, and therefore major changes to accounting practice where necessary to correct the problem.

Ball and Brown viewed this literature as providing a testable hypothesis – market participants should not be able to use earnings in a profitable manner.  Stated another way, knowing the amount of earnings that would be reported at the end of the year with certainty could not be used to profitably trade common stocks at the beginning of the year.  Evidence to the contrary would suggest the null that earnings are non-informative does not hold.

While the methods part of the paper is probably difficult for recent accounting archivalists to follow, Ball and Brown produce perhaps the single most famous graph in the accounting literature.  It shows stock returns trending up over the year for companies that ultimately report increases in earnings and trending down for companies that report decreases in earnings.  Thus they show that accounting numbers can be informative even if the aggregate number is not computed using a single unified measurement approach across transactions/events.  Subsequent research would show that numbers from the income statement have predictive ability for future earnings and cash flows.

As I sat listening to these two research icons, I could not help but think about some comments I have heard recently from a few standard setters and practitioners.  Those individuals express contempt for EPS in a mixed attribute world.  They appear to wish they could jump in a time machine and eliminate per share computations related to income. I readily admit that EPS does not explain much of the variance in returns over periods of one year or less ( e.g., Lev, JAR 1989). However the link is clearly significant, and over longer periods, the R2’s are quite high (Easton, Harris, and Ohlson, JAE 1992).  Can the standard setters make incremental improvements to increase usefulness of EPS?  I sure hope so, and maybe the recent paper posted by Alex Milburn will help.  But dismissing a reported number because it is not derived from a single consistent measurement attribute – be it fair value or historical cost – seems to revert back to pre-Ball and Brown views that are rejected by years of research.